I would like to make a little survey regarding the (geo)spatial projection that you use when elaborating your GPS and movement data for the spatial awareness of your robots.
Moving all GPS coordinates to a planar projection seems to be the more reasonable choice since not only distances (for which several formulas and approximations exist), but bearings must be computed.
Generally, although scales are pretty small here, avoiding equirectangular approximation seems a good idea in order to keep a more consistent system.
Avoiding working in the 3D world (Haversine and other great-circle stuff) is probably a good idea too to keep computations low-cost.
Moving the world to a 2D projection is hence what seems to be the best solution, despite reprojection of all input GPS coordinates is needed.
I would like to get opinions and ideas on the subject (...if ever anyone is interested in doing it U_U).